Flow-down sections imply flabbiness
Let \(R\) be a commutative ring whose multiplicative monoid has the DecompositionMonoid structure used in Lean. Its product-divisibility rule says: if finitely many scalars are pairwise relatively prime and each divides \(d\), their product divides \(d\). Polynomial rings over a field satisfy this rule.
Let \(G\) have finite vertex and edge sets, and let \(\mathcal O_c\) be its structure sheaf with labels \(\alpha_e\). Assume:
- the labels in each \(\operatorname{Out}_c(x)\) are pairwise relatively prime;
- a flow-down section exists at every \(x\le c\).
Then \(\mathcal O_c\) is flabby.
Relation to the OpenAI notes
Same model properties, different justification.
OpenAI notes: The induced-model lemma justifies applying the character theorem to the restricted reflection-faithful realization and uses the dihedral character calculation to identify the rank-one model. loc:models.
Lean (primary): The direct construction first proves parity matching in arbitrary lower Bruhat intervals by a simple-reflection involution. In the dihedral application, orbit-spline factors supply the prescribed values and zeros; their product gives flow-down sections. A finite induction on prescribed upper sets proves flabbiness, while generation and quotient maps follow from the structure-sheaf construction. DihedralIntervals:616, DihedralIntervals:622, DihedralIntervals:626, SheafMorphisms:252, SheafMorphisms:288, DihedralIntervals:176, DihedralIntervals:251, DihedralIntervals:372.
Role in the proof: The Blueprint follows the parity, normal-form prescription, spline product and flow-down correction arguments. Injectivity of the reflection-preserving group map provides distinct factor data and pairwise relatively prime incident labels. These model properties feed the later local-ring decomposition.
Proof
Step 1. Induct on the prescribed upper set
Let \(U\) be an upper vertex set and \(f\) a compatible assignment on it. Since the vertex set is finite, use strong induction on the finite set \(U\).
For \(U=\varnothing\), the zero global section meets the empty agreement requirement.
For nonempty \(U\), choose a minimal vertex \(x\in U\). The set \(U'=U\setminus\{x\}\) is upper: if \(y\in U'\), \(y\le z\), then \(z\in U\); equality \(z=x\) would contradict minimality of \(x\) and \(y\ne x\).
Step 2. Extend the smaller prescribed set
Restrict \(f\) to \(U'\). The induction hypothesis gives a global section \(g\) agreeing with \(f\) on \(U'\).
If \(x\not\le c\), its stalk is zero, so \(f_x=g_x=0\). Then \(g\) already agrees on all of \(U\).
Assume henceforth \(x\le c\), so both values at \(x\) are scalars in \(R\).
Step 3. Divide the remaining error by each outgoing label
Take a supported outgoing edge \(e:x\to y\), with \(y\le c\). Since \(U\) is upper and \(x<y\), its target lies in \(U'\). Thus \(g_y=f_y\).
Compatibility of \(f\) on \(U\) and of \(g\) globally gives \[ \alpha_e\mid f_x-f_y,\qquad \alpha_e\mid g_x-g_y. \] Subtracting, with \(f_y=g_y\), gives \[ \alpha_e\mid f_x-g_x. \]
Step 4. Divide by the full supported product
This holds for every \(e\in\operatorname{Out}_c(x)\). Pairwise relative primality and the product-divisibility rule imply \[ P_c(x)\mid f_x-g_x. \] Choose \(r\in R\) with \[ f_x-g_x=P_c(x)r. \]
Step 5. Correct the missing value with a flow-down section
Choose a flow-down section \(\theta_x\), and set \[ h=g+r\theta_x. \] The section module is closed under addition and scalar multiplication, so \(h\) is global. At \(x\), \[ h_x=g_x+rP_c(x)=f_x. \]
For \(y\in U'\), minimality of \(x\) implies \(y\not\le x\): otherwise \(x\le y\le x\), forcing \(y=x\). Hence \(\theta_x(y)=0\), and \[ h_y=g_y=f_y. \] Thus \(h\) agrees with \(f\) on all of \(U\), completing the induction.
The correction takes place after the smaller upper set has been extended. The lower support of \(\theta_x\) is exactly what preserves its already prescribed values.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.MomentGraph.Sheaf.structureCompatible_iff: DihedralIntervals.lean:359.
OAI.KLInvariance.MomentGraph.Sheaf.flowDown_flabby: DihedralIntervals.lean:372.